I don't really disagree with anything you just said, but it feels like a response to a different thread. Nobody here is arguing that randos who approach univers

I don't really disagree with anything you just said, but it feels like a response to a different thread. Nobody here is arguing that randos who approach university people are generally bringing valuable stuff to the table, or arguing that uneducated laypeople are likely to successfully unify QM and GR, or solve P=NP, etc. The topic at hand (at least as I took it, and what I was trying to address) is no more than endorsing somebody to submit to arXiv. I'm not suggesting academics should be champing at the bit to offer to collaborate with people who walk in off the street, or spend large amounts of time reading unsolicited papers in detail, etc. An arXiv endorsement is really just saying "this is generally appropriate to be on arXiv". They're explicit in saying it's not peer-review, and doesn't even require reading the work-to-be-submitted in detail. There are amazing discoveries that are that are made by relatively obscure people, but having one that is both simple and correct is extremely rare. True. But let me just point out that somebody wanting to submit to arXiv doesn't necessarily have to be claiming an "amazing" discovery. In fact, a lot of papers that get put up on arXiv represent incremental progress in a very narrow area, at best. I mean, go back to the era when researchers in meta-heuristics would routinely pick "random natural phenomenon" use it as a metaphor for a new meta-heuristic, code something up, improve by 1% on some published benchmark and name their "new algorithm" something like "the gopher tunnel method" or whatever and get a paper published. Some of those papers almost certainly wound up on arXiv before being published in whatever journal or conference proceeding they eventually wound up in. Speaking for myself, I know that if I were cold emailing somebody in academia to ask them to endorse me, it would almost certainly be for CS.LG or CS.AI and it would probably be something along those lines. An incremental improvement or minor "breakthrough" in an AI area. It wouldn't be to claim a solution to the P=NP problem or anything like that. And I'd hope that somebody would be willing to help out.